Causeway — AI Agent Factory on Amazon Bedrock

Teams are racing to build AI agents on Amazon Bedrock — and most are doing it the hard way. Agents start life in notebooks and one-off scripts, with no promotion path from prototype to production, no rollback, model IDs and secrets hard-coded, and no per-agent view of what any of it costs. The result is a sprawl of un-versioned, ungoverned, cost-opaque agents that Security and Finance cannot see, let alone control.

Causeway brings agent development under a repeatable, governed factory model — the same way you already treat the rest of your AWS estate as code. It is delivered as Terraform and runs entirely in your own AWS account; PurePlay Cloud never holds your data or your keys.

PurePlay Cloud's Causeway engagement is a fixed-scope, fixed-fee delivery that stands up the Agent Factory in your target AWS account, ships your first production agent end-to-end, and equips your platform team to build and operate every agent that follows — without ongoing dependency on us.

What we deliver

• Causeway deployed into your AWS account as Terraform — a shared platform stack plus your first per-agent stack, governed by your existing identity model.
• The factory's packaging model live in your account: every agent is an immutable, signed, digest-pinned container image plus a single agent.yaml manifest. Promotion dev→test→prod is a one-line version pin; rollback is the same change in reverse.
• Swappable runtimes from one container contract — Amazon Bedrock AgentCore Runtime by default (serverless, session-isolated), or ECS/EC2 by changing one manifest field; identical contract throughout.
• Per-agent FinOps, enforced not by convention but by IAM: each agent gets its own Bedrock application inference profile, a uniform pp:* cost-allocation tag set on every resource, AWS Budgets with alert thresholds, account-level Cost Anomaly Detection scoped to Bedrock, and CUR 2.0 → Amazon Athena showback. Every dollar of model spend is attributable to a named agent.
• Security posture as standard: per-agent least-privilege IAM execution roles, Bedrock Guardrails applied at every model call, and AgentCore Identity's token vault so agent code never sees long-lived secrets.
• Observability by default: OTEL traces into AgentCore / Amazon CloudWatch, a per-agent dashboard (tokens, invocations, latency, errors, guardrail interventions) and alarms wired to your chosen channel. No agent ships dark.
• Your first production agent, built end-to-end from a real use case and wired to the connectors it needs — choose from the catalog: Slack, Jira, Email (Amazon SES), Database, SharePoint, Outlook.
• Train-the-trainer for your platform team (typically a tech lead plus one or two engineers) so they can author manifests, add connectors, and promote releases without us.
• Architecture documentation and a recorded technical walkthrough handed to your team.

How we deliver it 10 PurePlay Cloud effort days over two calendar weeks, in four phases: Discovery (use cases, AWS topology, identity source, compliance drivers, and confirming runtime fit — AgentCore vs. ECS/EC2); Platform Enablement (Terraform platform stack, OIDC build/deploy roles, CI gates, FinOps and observability bootstraps); First Agent and Train-the-Trainer (manifest authoring, connector wiring, build→sign→promote, side-by-side with your trainer cohort and end-to-end validation); and Governance and Handover (Budgets, Cost Anomaly Detection, per-agent dashboards, onboarding runbook, documentation, recorded walkthrough).

Why PurePlay Cloud Our engineers came up building and operating AWS platforms for Australian public sector and regulated-industry customers — environments where IAM least privilege, CloudTrail forensics and cost control are not optional extras. Causeway is the productised form of those patterns, applied to AI agents on Bedrock. We run agents on this exact platform inside our own AWS account, so the engagement reflects a motion we have already walked, not theory from a vendor deck. Where AWS funding (MAP, POC) is available we pursue it on your behalf to offset the engagement cost.

Getting started Place an order to be contacted by our Client Engagement team within one business day. We will run a 30-minute qualification call, confirm AWS funding eligibility, and issue a tailored Statement of Work within five business days. Delivery typically commences within two weeks of signature.